Behavioral Indicators of Loneliness: Predicting University Students' Loneliness Scores from Smartphone Sensing Data

TL;DR

The paper tackles predicting university students' loneliness using passive smartphone sensing, addressing the dynamic nature of loneliness beyond retrospective self-reports. It combines generalized Random Forests with personalized Large Language Model inferences to predict ULS-8 scores from multi-sensor data collected over a semester from 88 students. Generalized RF models achieve MAEs of 3.29 (midterm) and 3.98 (end), with screen usage and location mobility as key predictors, while one-shot LLM personalization reduces prediction errors substantially (up to ~42%), highlighting the value of contextual, individual-level signals. The work demonstrates the feasibility of scalable, interpretable digital phenotyping for loneliness with potential applications in digital mental health of college students, while noting limitations related to data diversity and contextual interpretation.

Abstract

Loneliness is a critical mental health issue among university students, yet traditional monitoring methods rely primarily on retrospective self-reports and often lack real-time behavioral context. This study explores the use of passive smartphone sensing data to predict loneliness levels, addressing the limitations of existing approaches in capturing its dynamic nature. We integrate smartphone sensing with machine learning and large language models respectively to develop generalized and personalized models. Our Random Forest generalized models achieved mean absolute errors of 3.29 at midterm and 3.98 (out of 32) at the end of semester on the UCLA Loneliness Scale (short form), identifying smartphone screen usage and location mobility to be key predictors. The one-shot approach leveraging large language models reduced prediction errors by up to 42% compared to zero-shot inference. The one-shot results from personalized models highlighted screen usage, application usage, battery, and location transitions as salient behavioral indicators. These findings demonstrate the potential of smartphone sensing data for scalable and interpretable loneliness detection in digital mental health.

Figures (6)

• Figure 1: Zero-shot prompt instruction for LLMs to predict the ULS-8 loneliness score using smartphone usage data.
• Figure 2: One-shot prompt instruction for LLMs to predict the ULS-8 loneliness score using smartphone sensor data.
• Figure 3: Comparison of feature extraction errors across smartphone sensors using LLMs, averaged by participants’ total ULS-8 loneliness scores (N = 88).
• Figure 4: Mean Absolute Error (MAE) of zero-shot and one-shot inferences for predicting ULS-8 item scores across different smartphone sensors.
• Figure 5: Dot plot distribution of remaining feature count from smartphone sensors vs. Cross-Validation MAE in ULS-8 Loneliness prediction models at midterm (left) and end-of-semester (right).